Breaking the Capability Ceiling of LLM Post-Training by Reintroducing Markov States
This addresses a fundamental bottleneck in aligning LLMs for open-ended discovery, offering a novel approach to enhance reasoning capabilities beyond incremental refinements.
The paper tackles the 'capability ceiling' in RL-based post-training of LLMs, where current methods merely refine pre-existing patterns, by reintroducing explicit Markov states to reduce sample complexity and break performance boundaries, achieving consistent gains in complex logic puzzles.
Reinforcement learning (RL) has become a standard paradigm for post-training and aligning Large Language Models (LLMs), yet recent evidence suggests it faces a persistent "capability ceiling": unlike classical RL systems that discover novel strategies, RL for LLMs often acts as a mere refiner of patterns already latent in pre-trained weights. In this work, we identify a fundamental structural bottleneck: while classical RL relies on compact, informative Markov states, current LLM post-training formulations are tethered to an ever-expanding history of actions. We revisit a classical principle long central to RL yet absent from LLM post-training: explicit Markov states. Theoretically, we provide rigorous guarantees demonstrating that leveraging estimated Markov states can significantly reduce sample complexity. Empirically, we show that introducing Markov states consistently breaks the performance boundaries of standard RL post-training across a suite of complex logic puzzles. Our findings suggest that moving beyond "history-as-state" modeling in favor of structured Markovian representations is essential for unlocking open-ended discovery and genuinely new reasoning capabilities in Generative AI.